Count Models of Social Networks in Finance

Abstract

We use overdispersed Poisson regression models to study social networks in finance. We count an investor's social connections in different cities as proportional to the number of stocks held by this investor that are headquartered in those cities. When connections are formed in an i.i.d. manner, the count of such connections in any city follows a Poisson distribution. Using data from institutional investors' holdings, we find instead overdispersion for some cities like San Jose and San Diego, which suggests that investors have non-i.i.d. propensities to be connected to these cities. Indeed, these overdispersed cities have a large number of graduates from local universities who work in the fund industry. Managers with non-i.i.d. propensities to pick stocks from these overdispersed cities significantly outperform other managers. Our effect is not driven by industry concentration or familiarity bias of managers.(With Jiangmin Xu)